A hybrid fault isolation method for acceleration sensors in maglev train using auto-regression model and improved extended state observer

In this paper, a hybrid method is proposed to solve the problem of dual acceleration sensor fault isolation in maglev suspension system. Firstly, considering uncertain external disturbance and time-varying measurement noise of the suspension system, the nonlinear dynamic model of the suspension system is established. Secondly, an improved Sage-Husa adaptive Kalman filter method is adopted to extract the position differential signal under the background of time-varying noise for real-time estimation of track irregularity. AR model is used to model track irregularity, and a prediction phase is carried out to reduce the dimension of unknown disturbance after the occurrence of fault. Then, considering the nonlinearity of the suspension model, the unknown external force disturbance and the sensor measurement noise, an improved extended state observer (ESO) is imported to generate the reference velocity signal utilizing the position signal. Sensor fault isolation is achieved by residual evaluation. Finally, the effectiveness of this hybrid fault isolation method is verified by simulation.